Working Thesis
Teresa Rokos
February 2019
Abstract
This thesis evaluates how the market power of insurers and hospitals in geographical sub-markets determines the prices insurers and hospitals negotiate for hospital care. I will examine to what extent the variation in prices is explained by variation in costs and quality versus variation in market power, which may play an important role in bilateral price negotiations for services. Using carrier-hospital specific prices for outpatient medical procedures from New Hampshire’s NH HealthCost website, I compare the estimated prices for the same services across the state, both between hospitals and within hospitals. The methodology for the collection of this data can be found here.
Introduction
Motivation
The U.S.’s expenditure on health care has grown rapidly over the last few decades and has outpaced that of many peer countries without significant returns to quality (e.g. OECD 2017; Etehad and Kim 2017; Reinhardt et al., 2004). A substantial body of research has determined that the biggest driver behind the U.S. health care expenditure is not greater utilization or social spending but quite simply higher prices (e.g. Anderson et al., 2003; Papanicolas et al., 2018). Given that the $3.3 trillion spent on health care annually (17.9% of GDP) (Centers for Disease Control and Prevention 2017) erodes the budgets of families and individuals, state and local governments, and the federal government, understanding the sources of variation in these prices is of tremendous import.
I choose to focus this paper on hospital prices because hospital care constitutes the largest component of U.S. health care spending at 32.4% ($1.0825 trillion) of national health expenditure (Centers for Disease Control and Prevention 2017) as shown in Figure 1, and a vast body of research has identified wide variation in private insurance reimbursements to hospitals for the same services and/or diagnoses citations. Cooper et al., 2015 finds that for the privately insured, only about half of the variation in expenditure is due to the quantity of care delivered while half is purely driven by price variation; for public programs like Medicare, 95% of the variation in expenditure is explained by variation in the quantity of care delivered. 
Given that the largest portion of Americans (49%) receive health care coverage from their employer through private insurance plans rather than through Medicare and Medicaid (Kaiser Family Foundation 2017), this price variation among private insurers may affect what many Americans pay out of pocket for their healthcare. Moreover, this prior research such as Cooper et al. 2015 finds that this variation persists even when costs and quality are held constant.
This paper is most closely related to Cooper et al. 2015, but differs in several key respects: first, this analysis will focus a set of price data for hospital care in New Hampshire, while Cooper et al. 2015 relies on HCCI data which is a national cross-section sample and relies on claims data from Aetna, Humana, and United Health. While the HCCI data is a rich set of data that is perhaps nationally representative in a broad sense, none of these insurers are major players in the New Hampshire market; secondly, this paper will devote more attention to insurers’ relative market power as well as the effects on the uninsured who have minimal market power; lastly, I also include an analysis of how the share of Medicare patients at a given hospital affects the prices that hospital negotiates with private insurers.
Theoretical Framework
In standard micro-economic theory, one learns that prices reflect supply and demand, where the supply curve is largely determined by each firm’s costs of production and market power and the demand curve determined by consumer preferences and budget constraints. Market power is largely a function of the number of other firms in the market, but a firm may also generate market power for itself through distiguishing it’s product from others, most commonly in the form of quality. When it’s product differs from others in the market, a firm can extract a certain mark-up from consumers who have a preference for that difference, and consumers often have a preference for higher quality. Thus, in a standard competitive market, one would expect prices to reflect production costs and product quality.
The market for health services (often provided through hospitals) is unique in that there are two components of demand: individuals and payers. Even this is somewhat of an oversimplification in the US market because many individuals are privately insured through their employer which contracts with a private insurer/payer. However, for the most part, insurers cover much of the costs of care citation and negotiate contracts with providers if they are private (Medicare and Medicaid set their prices at a regional level). This gives private payers some form of monopsony power as there are often only a few of them present in each market, and they can choose whether or not to include each provider in their network which affects the prices individuals face for care.
Thus, payers and providers engage in bilateral negotiations where both are price setters rather than price takers. I assume the following: both the provider and the payer want to maximize their profit margins per patient, even if the provider is a non-profit (Horwitz 2005), and the number of patients they serve. Providers want to receive higher prices for their services but also to be included in payers’ networks, as patients face lower out-of-pocket costs when going to providers in network. Payers want to pay lower prices for providers’ services to minimize their costs but also want to attract enrollees by including convenient providers in their network. Thus, this is the bargaining environment in which bilateral hospital-insurer contracts are negotiated and the work below examines the impact on prices.
Background on the New Hampshire Market
Health Insurance Coverage
New Hampshire is a small state (8,953 sqaure miles) with a population of 1.343 million (90.5% white) and median household income of $68,485 (Census Bureau 2017). The following pie charts depict the insurance status of New Hampshire residents in 2016 as well as the market share of major private insurers. According to the New Hampshire Insurance Department, a total of 543,900 members were enrolled in commercial market plans in 2017 in the state, and 81% received coverage through employer-sponsored insurance plans (11.0% Individual Market, 7.7% NH PAP, 12.6% Small Group, 19.8% Large Group Fully Insured, and 48.9% Large Group Self Insured).
New Hampshire has a total of 31 hospitals and 3,503 beds (New Hampshire Hospital Association 2018). According to the New Hampshire Hospital Association, 13 of these hospitals with 2,704 beds total are Prospective Payment Systems (PPS) Hospitals, which take prospective payments from Medicare; 13 hospitals with 301 beds total are Critical Access Hospitals (CAH), which serve rural populations and therefore are eligible to receive certain benefits from Medicare such as cost-based reimbursement for Medicare services (Rural Health Information Hub 2018); 5 hospitals with 498 beds total are Specialty Hospitals, which provide a specialized category of services (such as children’s hospitals, orthopedic hospitals, cancer hospitals, etc.). All of these hospitals can serve non-Medicare patients as well.
Location of New Hampshire Hospitals (from Google Maps data):
Price Variation in the New Hampshire Market
I first want to motivate this paper by demonstrating that large price variation exists in the New Hampshire market for hospital care and that little of the variation is explained by costs and quality.
The New Hampshire Hospital Association provides a list of the top outpatient CPT codes in 2016 at hospitals throughout the state, and the average costs associated with each. Current Procedural Terminology (CPT) codes provide a uniform language for coding medical procedures and services and are published and maintained by the American Medical Association (AMA). The New Hampshire Hospital Association lists the 82 most common outpatient CPT codes (out of several thousand). To provide a picture of the extent of variation in the prices for these services, I further narrow this list down to 10 of the most expensive services for which I can obtain data on private insurance payments and prices for uninsured individuals from NH HealthCost and Medicare payments from CMS.
The Medicare prices were constructed for the relevant procedures/CPT codes using data from the 2018 Physician Fee Schedule (PFS) Relative Value Files and the following formula:
The 2018 Medicare Facility Pricing Amount is constructed for each CPT code (each procedure) at each hospital/facility from the 2018 Relative Value Files available through CMS. I then combine these Medicare prices with those available through NH Health Cost. For ten common procedures, I present the resulting boxplot distributions in Figure 2.1, which includes payments by the uninsured, private insurers, and Medicare, and in Figure 2.2, which only shows the payments made by private insurers. All price data presented reflect 2018 prices.
Clearly, significant variation exists, but I now want to examine how this variation relates to costs and quality. In my model, I find that, controlling for procedure, across hospital price variation accounts for 16.4% of the total variation in the model and within hospital price variation accounts for a further 17.0% of the total variation (using ANOVA). If prices truly reflected costs and quality, one may expect to see prices vary across hospitals which may have varying levels of quality, but within the same hospital, one would expect both price and quality to be constant for the same procedure across patients with different insurers.
I first compare prices across hospitals using measures for overall quality obtained through CMS Hospital Compare data. The hospital overall ratings show the quality of care a hospital may provide compared to other hospitals based on the quality measures reported on Hospital Compare, summarizing into a single rating more than 60 measures in seven measure groups: mortality, safety of care, readmission, patient experience, effectiveness of care, timeliness of care, and efficient use of medical imaging. I do not find any statistically significant nor visually apparent relationship between CMS’s Overall Hospital Rating and the prices charged to private payers at that hospital; if anything, the relationship appears slightly negative. 
| Dependent variable: | |
| Percent of Medicare Price | |
| Overall Hospital Rating (Hospital Quality) | 0.312 |
| (0.348) | |
| Observations | 3,652 |
| R2 | 0.536 |
| Adjusted R2 | 0.531 |
| Residual Std. Error | 15.704 (df = 3612) |
| F Statistic | 106.845*** (df = 39; 3612) |
| Note: | p<0.1; p<0.05; p<0.01 |
However, I do find that for the exact same service, hospitals are able to charge different payers vastly different amounts. Below are two visual examples from the data that illustrate this variation between the three major private payers in New Hampshire (Anthem NH, CIGNA, and Harvard Pilgrim Health Care) for the same service at the same hospital.
These graphs present several examples of striking price variation for two common radiology services, Ultrasound - Tranvaginal (non-maternity) and X-Ray - Spine (outpatient). For all of the hospitals in the ultrasound graph, there is a difference of at least $150 in the price paid by the highest payer and that paid by the lowest payer. In three of these hospitals (St. Joseph Hospital, Anna Jaques Hospital, and Foundation Medical Partners), the highest payer is paying more than twice the amount the lowest payer is paying. For all of the hospitals in the x-ray graph, there is a difference of at least $100 between the highest and lowest payers. I make the assumption that it is unlikely that the cost of providing each of these services and the quality of each of these services within the same hospital should not vary with patients’ insurance coverage. That is, it is reasonable to assume that it is no more costly from the hospital’s point of view to perform a spine x-ray on a patient covered by CIGNA as opposed to a patient covered by Anthem, and hospitals do not provide higher quality spine x-rays to CIGNA covered patients that they do to Anthem patients.
Seeing that hospital price variation does not appear greatly correlated with costs or quality within hospitals, the analysis that follows will investigate how the market power of the insurer and the market power of the hospital relate to price.
Data Sources
Raw data from NH HealthCost
The primary source of data on hospital costs comes from NH HealthCost which is a publicly available tool that provides insurer-hospital specific estimations of total costs (reimbursements) for a long list of outpatient services and procedures. All cost information is based on claims data collected in the New Hampshire Comprehensive Healthcare Information System which unfortunately I was unable to access directly. On the NH HealthCost website, the average amount each insurer pays for a given service at a given hospital is available, and it is reasonable to assume that this is roughly equivalent to the price each hospital and insurerance carrier have negotiated for each service. This allowed me to construct a data set containing the estimated price for each procedure and each provider-insurer combination for which data was available.
CMS Data
The Center for Medicaid and Medicare Services (CMS) provides several useful data sets that aid my analysis. The first is the Physician Fee Schedule (PFS) Relative Value files which allow me to calculate the Medicare reimbursement rates at each hospital for each service observed in the NH HealthCost data. This provides me with a baseline Medicare price for each service for each locality, and Medicare treats all of New Hampshire as one locality, meaning that all hospitals in New Hampshire are reimbursed the same amount for outpatient services. this may not be true for Critical Access Hospitals (CAHs) of which New Hampshire has 13 so I need to figure out at what rate Medicare reimburses these hospitals This allows me to standardize and compare how much above this baseline rate private insurers or the uninsured pay.The second dataset provided by CMS is the Hospital Compare data which provides measures of quality.
Methodology
There are two stages to my analysis, and in both I choose to restrict my investigation to radiology services as these types of services (CT scans, MRIs, X-rays, etc.) can reasonably be assumed to have constant costs and quality within the same hospital. The first stage seeks to determine whether there is a state-wide statistical difference between the prices paid by the three major private insurers and the second stage evaluates whether hospitals with greater market power (defined as market share) extract higher prices.
First Stage: Identifying Price Variation Patterns Across Major Private Insurers
The goal of this analysis is to determine whether there is a statistical difference between the prices hospitals charge Anthem, Harvard Pilgrim, and CIGNA in New Hampshire. That is, for the same service at the same hospital, is one insurer consistently paying a higher amount than the other two?
I create dummy variables for each of the insurers and Medicare and a new variable (“percent_medicare”) which is the price as a percentage of what Medicare pays for the same procedure at the same hospital. The price Medicare pays for Blood Tests is $0 so I drop those observations from my data (from 6,329 to 6,002). With the remaining data, I regress the “percent_medicare” variable on each insurer dummy variable (with the Medicare dummy being excluded) and include procedure fixed effects and hospital fixed effects to isolate within-hospital variation of prices and therefore plausibly control for quality and costs.
The following boxplot depicts the various distributions of estimated prices for each payer-type across all 32 radiology services available on NH HealthCost. 
The
Second Stage: Comparing Price Variation and Hospital Concentration
For now, I restrict this second stage regression to only hospitals included in the Hospital Compare data set as these are hospitals that Medicare officially recognizes as Acute Care or Critical Access Hospitals. I then compare how the prices for each insurer varies with how many hospitals are within each hospital’s market. I define a hospital’s market as being within +/- .2 degrees of latitude and longitude of that hospital. I also include the median household income of the county obtained from 2017 US Census Data.
| Dependent variable: | |
| Percent of Medicare Price | |
| Number of Hospitals in Market | 1.222*** |
| (0.370) | |
| Median Household Income (county) | 0.00000 |
| (0.00001) | |
| Overall Hospital Rating - 3 | 2.791*** |
| (0.600) | |
| Overall Hospital Rating - 4 | 2.213*** |
| (0.582) | |
| Overall Hospital Rating - 5 | -0.289 |
| (0.756) | |
| Constant | 157.592*** |
| (1.776) | |
| Observations | 1,443 |
| R2 | 0.904 |
| Adjusted R2 | 0.901 |
| Residual Std. Error | 4.912 (df = 1403) |
| F Statistic | 337.663*** (df = 39; 1403) |
| Note: | p<0.1; p<0.05; p<0.01 |
Robustness
Results
Price Variation Patterns Across Major Private Insurers
The results are listed below and show that Anthem pays on average 887% of Medicare prices, Harvard Pilgrim pays 920%, and CIGNA pays 1049%.
| Dependent variable: | |
| Percent of Medicare Price | |
| Anthem NH | 8.878*** |
| (0.434) | |
| Harvard Pilgrim HC | 9.200*** |
| (0.389) | |
| CIGNA | 10.469*** |
| (0.448) | |
| Observations | 3,814 |
| R2 | 0.565 |
| Adjusted R2 | 0.545 |
| Residual Std. Error | 9.265 (df = 3642) |
| F Statistic | 27.717*** (df = 171; 3642) |
| Note: | p<0.1; p<0.05; p<0.01 |